Upload
others
View
10
Download
0
Embed Size (px)
Citation preview
ORI GIN AL PA PER
Improving representativeness of genebank collectionsthrough species distribution models, gap analysisand ecogeographical maps
M. Parra-Quijano • J. M. Iriondo • E. Torres
Received: 27 April 2011 / Accepted: 3 October 2011 / Published online: 16 October 2011� Springer Science+Business Media B.V. 2011
Abstract An efficient germplasm collecting method was evaluated using six Lupinusspecies and the Spanish Lupinus collection as a study case. This method includes the
application of geographic information systems, ecogeographical land characterization
maps, species distribution models and gap analysis to identify prioritized collecting sites.
To evaluate the efficiency of this collecting method, field collecting expeditions were
carried out focusing on prioritized sites and the results of these collections were analyzed.
Prioritized sites were identified using spatial and ecogeographical gaps, and potential
species richness maps. The spatial gaps corresponded to populations non-included in the
collection but recorded by other information sources while ecogeographical gaps corre-
sponded to spatial gaps that were located in ecogeographical categories (obtained from the
ecogeographical map) that were scarcely represented in the collection. A potential Lupinusspecies richness map was obtained by adding the information of single maps of Lupinusspecies distribution models. Subsequently, prioritized sites were obtained in ecogeo-
graphical gaps with high potential species richness values. Collecting expeditions were
made in Spain in 2006, 2007 and 2008. Results showed that using the efficient germplasm
collecting methodology was highly positive not only from a quantitative viewpoint
(between 7.8 and 11% increase) but also in qualitative terms, focusing collection efforts
in ecogeographical categories with low or null representation in the Spanish Lupinuscollection (41% of the new accessions). Phenotypic differences related to adaptation to
environment were observed in the field between the populations that grow in low or null
represented categories and those that grow in highly represented categories.
M. Parra-QuijanoFacultad de Agronomıa, Universidad Nacional de Colombia sede Bogota,Apartado Aereo, 14490 Bogota, Colombia
M. Parra-Quijano (&) � E. TorresDepartamento de Biologıa Vegetal, Universidad Politecnica de Madrid,Ciudad Universitaria s/n, 28040 Madrid, Spaine-mail: [email protected]; [email protected]
J. M. IriondoArea de Biodiversidad y Conservacion, Universidad Rey Juan Carlos,Tulipan s/n 28933 Mostoles, Madrid, Spain
123
Biodivers Conserv (2012) 21:79–96DOI 10.1007/s10531-011-0167-0
Keywords Agrobiodiversity � Collecting indices � Efficient germplasm collection �Genebank representativeness � Lupinus
Introduction
The number of gene banks has increased steadily since they were first established in
the 1920’s. According to the second report on The State of the World’s Plant Genetic
Resources for Food and Agriculture, there are now some 1,750 gene banks worldwide, with
about 130 of them each holding more than 10,000 accessions (FAO 2010). As the number
of accessions and wild or crop species included in these gene banks increases, goals for
plant genetic resources are shifting from a purely quantitative perspective to a new par-
adigm that focuses on the quality of collections. In this sense, collections are expected to
be representative of the overall existing genetic variation present in nature or in the field
across the distribution of the target taxa. Since ex situ collections aim to cover the max-
imum amount of genetic variation and the entire range of environmental adaptation of the
target species, nowadays the objective in collecting expeditions is frequently to fill gaps in
representativeness. The origin of these gaps or biases in plant genetic resources collecting
is described in detail by Hijmans et al. (2000).
The representativeness of ex situ collections can be measured in genetic or ecogeo-
graphical terms (Parra-Quijano et al. 2008). Both of these intrinsically-related aspects can
be assessed by comparing total and sampled genetic diversity or ecogeographical range.
Morphological descriptors, molecular markers or agronomic evaluation data can be used to
assess genetic diversity in collections. However, estimation of genetic representativeness
(GR) requires the assessment of the genetic diversity of the existing populations in nature
or in the field, which can be an unattainable task. Furthermore, GR studies require
expensive infrastructure and highly qualified personnel, making its application difficult in
developing countries or in large ex situ collections. Alternatively, ecogeographical rep-
resentativeness (ER) can indirectly reflect GR, due to the relationship that exists between
the environmental characteristics of a site and the genetic features of the populations
occurring at that site through natural selection and local adaptation (Greene and Hart
1999). Thus, ER can be useful in estimating the GR of germplasm collections using simple
and cost-effective methods. In this sense, the increasing availability of geographic infor-
mation systems (GIS) software and environmental datasets facilitate ER assessment
(Guarino et al. 2002).
ER analysis can be used to detect ecogeographical gaps in ex situ collections and
subsequently identify where to prioritize collection efforts. Advances in ecogeographical
land characterization (ELC) maps (Parra-Quijano et al. 2008), gap analysis (Maxted et al.
2008) and species distribution models (Jarvis et al. 2005) make it possible to detect such
ecogeographical gaps and locate potential collecting sites.
Optimized collecting strategies for existing germplasm collections aim to maximize the
GR of collection while minimizing the economic cost of collecting missions. To achieve
this, selected sites for germplasm collection must meet the following conditions: (1) the
site must have a high probability of occurrence of the target species, (2) the site must not be
represented in the existing ex situ collection, and (3) the location must have environmental
conditions that are under-represented or even non-represented in the existing collection.
Lupinus species have a high potential as a protein source crop (Sujak et al. 2006), and
exceptional adaptation to extreme biotic and abiotic conditions (Rahman and Gladstones
80 Biodivers Conserv (2012) 21:79–96
123
1974; Gladstones 1980; Hamblin et al. 1986; Huyghe 1997; French et al. 2001). The Plant
Genetic Resources Centre of the National Institute of Agriculture and Food Research
(CRF-INIA) holds one of the most important Lupinus germplasm collections in the world,
with more than 2,100 accessions of wild and cultivated species. Because the Iberian
Peninsula has highly heterogeneous environmental conditions, the aim of this study was to
assess the effectiveness of optimized collection in improving the ER of the CRF-INIA
Lupinus collection. The development of optimized collection involved the application
of GIS tools, gap analysis, species distribution models and ELC maps. The generated
methodology was put into practice and the results of the collecting expeditions were
evaluated both in quantitative and qualitative terms.
Materials and methods
Lupinus species presence data
CRF-INIA collection dataset
The CRF-INIA Lupinus collection contains 1,026 georeferenced accessions collected
in Peninsular Spain and the Balearic Islands, including six species: Lupinus albus (248
landraces), Lupinus angustifolius (489 wild populations), Lupinus cosentinii (5 wild pop-
ulations), Lupinus hispanicus (175 wild populations), Lupinus luteus (6 landraces, 86 wild
and 12 weed populations) and Lupinus micranthus (5 wild populations).
External dataset
To assess the representativeness of the CRF-INIA Lupinus collection, information on the
total occurrence of Lupinus spp. in the Iberian Peninsula and Balearic Islands was gathered
from herbarium vouchers preserved at the Royal Botanical Garden of Madrid (MA, Spain),
the ANTHOS project database (http://www.anthos.es) and the Global Biodiversity Infor-
mation Facility (GBIF) database (http://data.gbif.org).
In total 280 presence data for five species were compiled: 29 corresponded to landraces
of L. albus, 135 to wild populations of L. angustifolius, 65 to wild populations of L. hispa-nicus, 38 to wild populations of L. luteus and 13 to wild populations of L. micranthus.
Point maps
CRF-INIA and external presence data were spatially represented as point maps using
DIVA-GIS (Hijmans et al. 2001). These maps and their associated tables were used to
check the geographical integrity of the coordinates following Hijmans et al. (1999).
Spatial gap detection
Circular buffer zones with a 1 and 10 km radius were created around the CRF-INIA data
points. Circular buffer zones with a 1 km radius were also created around the external data
points. The position of the different buffer zones could result in three possible situations:
(1) the external data buffer zones intersect the CRF-INIA data buffer zones with a 1 km
radius, (2) the external data buffer zones only intersect the CRF-INIA data buffer zones
with a 10 km radius, and (3) the external data buffer zones do not intersect any of the
Biodivers Conserv (2012) 21:79–96 81
123
CRF-INIA data buffer zones (those with a 1 km or 10 km radius). The first case was
considered to indicate ‘‘no gap’’ as it would imply collecting germplasm in populations
located less than 2 km from those represented in the CRF-INIA, while the second and third
case were considered to indicate mid-priority and high-priority spatial gaps, respectively
(Fig. 1). Maps containing mid and high-priority spatial gaps were created for each Lupinusspecies.
Ecogeographical gap detection
ELC map
A previously developed ELC map for Peninsular Spain and the Balearic Islands (Parra-
Quijano et al. 2011) was used to detect ecogeographical gaps in the collection. The ELC
map reflects the existence of different abiotic environments (or ecogeographical catego-
ries) in the territory under study. The ecogeographical categories were defined through
multivariate analysis of 61 environmental variables grouped into three modules: climatic,
edaphic and geophysical. The module of climatic variables included annual and monthly
precipitation, monthly mean, minimum and maximum temperature and annual mean
temperature (Hijmans et al. 2005). The edaphic module was composed of USDA soil type
and mineral predominant material (Instituto Geografico Nacional 1992), and the geo-
physical module comprised altitude, longitude, latitude, slope and aspect [derived from
SRTM digital elevation model, see Rabus et al. (2003)]. Each group was subjected to two-
step cluster analysis (SPSS 2003) automatically producing three clusters for each module.
Finally, the combination of clusters provided 27 unique ecogeographical categories
(3 climatic 9 3 edaphic 9 3 geophysical).
The ELC map was obtained by assigning one ecogeographical category to each of the
764,673 1 9 1 km cells included in the work space. This map has been positively tested
for detection of adaptation traits in several legume species, including L. angustifolius(Parra-Quijano et al. 2011).
Extraction of ecogeographical categories
Ecogeographical categories corresponding to the presence points of Lupinus species were
extracted by overlapping the point maps (CRF-INIA and external presence data) on the
ELC map using DIVA-GIS (Hijmans et al. 2001). Thus, one ecogeographical category was
assigned to each point according to its location. Frequency distributions, for each species
and data set, were obtained from the number of points falling into each ecogeographical
category.
Fig. 1 Detection of spatial gaps. a No gap, b mid-priority gap and c high-priority gap
82 Biodivers Conserv (2012) 21:79–96
123
Ecogeographical gaps in the CRF-INIA collection
The frequency distributions obtained for the CRF-INIA data set were used to classify the
ecogeographical categories into quartiles as follows: low (25% of ecogeographical cate-
gories with the lowest frequency, 0–25 percentile), mid-low (from 25 to 50 percentile),
mid-high (from 50 to 75 percentile) and high frequency classes (25% of ecogeographical
categories with the highest frequency–75–100 percentile). This classification was carried
out individually for the six Lupinus species considered. External presence data classified as
mid or high-priority spatial gaps were also considered ecogeographical gaps if their eco-
geographical category is classified as low or mid-low category frequency, according to the
previous classification.
Potential areas of high Lupinus species richness
General lineal model (GLM) procedure was used to predict areas of high Lupinus species
richness in the Iberian Peninsula and Balearic Islands. The distribution of four Lupinusspecies was modeled (L. albus, L. angustifolius, L. hispanicus and L. luteus) while
L. cosentinii and L. micranthus models were not performed because the number of pres-
ence data for these species was insufficient. Species distribution models were obtained
following Parra-Quijano et al. (2007) with some modifications: (a) only external data were
considered as presence (training) data to obtain the models, (b) pseudo-absences were
randomly chosen from areas belonging to the low and mid-low frequency classes according
to the classification of ecogeographical categories for each Lupinus species (see above),
(c) cell size was enlarged from 500 m to 1 km. Binary maps (0 = unsuitable, 1 = high
probability) were then produced for each species using a threshold of P C 0.8. A final map
of potential areas of species richness was obtained by the sum of the binary maps.
The performance of the species distribution models was validated by the Kappa statistic,
which was obtained from confusion matrices (Fielding and Bell 1997) using the popula-
tions detected in collecting explorations as testing data.
Selecting priority sites
Maps of ecogeographical and spatial gaps were superimposed on the map of potential
species richness. Matches between gaps and high potential species richness areas were
considered possible collection sites. The large number of potential sites obtained was
subjected to a prioritization process. Thus, a potential site was considered a priority site for
collecting Lupinus germplasm if it met each of the following criteria: it should represent a
mid or high-priority spatial gap, a low or mid-low frequency group of ecogeographical
categories (ecogeographical gaps) and a high probability area for one or more Lupinusspecies. Thus, a set of priority sites for collecting was defined for each Lupinus species and
projected on a map.
The selected priority sites for collecting, together with additional information (roads,
municipalities and land cover maps, and detailed satellite images), were loaded in the GIS
in order to define the most efficient collection routes.
Collecting Lupinus in priority sites
Collecting expeditions were carried out in the spring and summer from 2006 to 2008.
Explorations included selected priority sites and their surroundings. Surrounding areas
Biodivers Conserv (2012) 21:79–96 83
123
comprise a ring of a 5 km diameter around the priority sites. Only the roadsides were
examined in the surrounding areas. Populations of a particular Lupinus species found in
priority sites defined for other Lupinus species were also collected, as long as these pop-
ulations were not already represented in the CRF-INIA collection. Thus, populations
detected or collected for each species were classified as follow: (a) inside priority sites
(defined above) and (b) outside priority sites, including surrounding areas and priority sites
for other species.
All Lupinus populations found were georeferenced, and additional passport data were
gathered when seed collection was successful.
Evaluation of the improvement in ER
Quantitative and qualitative parameters were used to evaluate the real and theoretical
efficiency of the proposed collecting strategy.
Quantitative evaluation first considered the increase in the size of the CRF-INIA
collection. This increase was calculated using both the number of effectively collected
populations (real increase) and the number of detected populations (theoretical increase).
Other parameters used to evaluate the efficiency of the method were:
– Efficiency in population detection (EPD), defined as the ratio between the number of
populations detected in situ and the number of sites visited expressed in percentage.
EPD was calculated for both inside and outside selected priority sites (EPDi and EPDo,
respectively) for each Lupinus species.
– Efficiency in accession collection (EAC), defined as the ratio between the number of
populations collected and the number of populations detected in situ expressed in
percentage. Just a single overall EAC was calculated for each Lupinus species, because
once the population has been detected, the probability of collecting seeds is the same
regardless of whether the population is located inside or outside priority sites.
– General collecting index (GCI) or the total number of collected populations
(considering all species) divided by the total number of visited sites (considering
both inside and outside selected priority sites).
Qualitative evaluation shows the performance of the proposed methodology concerning
the collecting of populations that grow in low or non-represented environments of the
target germplasm collection. Therefore, the results of the expeditions were evaluated
taking into account the ecogeographical categories of the populations detected or collected
in relation to the CRF-INIA frequency classification previously described. Then, collecting
results were classified into non-represented, low, mid-low, mid-high and high frequency
ecogeographical categories classes. Thus, evaluation parameters, such as size increase and
efficiency indices used in the quantitative evaluation, were also considered in the quali-
tative evaluation, but they were calculated independently for each frequency class.
Results
Validation of the potential species richness map
Kappa values for each species model (L. albus = 0, L. angustifolius = 0.42, L. hispani-cus = 0.45 and L. luteus = 0.43) indicate poor fit for L. albus and a good fit for the rest of
the species models according to Landis and Kock (1977) criterion.
84 Biodivers Conserv (2012) 21:79–96
123
Spatial and ecogeographical gaps and priority sites for collecting
Eighty-two spatial gaps were detected, 52 of which were classified as high priority and 30
as mid-high priority. Lupinus angustifolius was the species for which the highest number of
gaps was detected (Table 1). As no external presence data were available for L. cosentinii,no gaps were detected for this species.
When spatial gaps were classified according to the categories from the ELC map,
79 ecogeographical gaps were obtained, most of which corresponded to the ‘‘mid-low
frequency’’ class. When the ecogeographical gaps where overlapped on the map of
potential species richness, 21 gaps were located in areas of non-potential occurrence of
Lupinus species and were discarded. The remaining ecogeographical gaps corresponded to
areas with up to three predicted species occurrence. Finally, 58 priority sites were defined
and mapped. The priority sites for collecting mostly corresponded to L. angustifolius,
L. hispanicus and L. luteus, while no priority sites were defined for L. albus and
L. cosentinii (Table 1). Figure 2 shows the process and the resulting maps of the different
steps taken to detect priority sites, using L. angustifolius as an example.
Detected and collected Lupinus populations
A total of 102 (1 9 1 km) sites were visited including selected priority sites and their
surroundings. In these areas, 114 populations of four Lupinus species were detected
(Table 2; Fig. 3b). Seeds were collected in 80 of these populations (Fig. 3c) and included
as new accessions in the CRF-INIA Lupinus collection.
Improvement in the representativeness of the CRF-INIA collection
Quantitative evaluation
The CRF-INIA Lupinus collection increased in size by 7.8%, but this increase would have
been 11.1% if all detected populations had been collected. In relative terms L. cosentiniiwas the species with the greatest increase in size (40%), while in absolute terms the
Table 1 Spatial and ecogeographical gaps and priority sites for collecting by Lupinus species
Species Spatialgaps
Ecogeographicalgaps
Ecogeographical gaps under differentpotential species richness areas
Number ofpriority sites
Number of species
MPa HPb L Fc M-L Fd 1 2 3
L. albus 0 2 0 1 0 0 0 0
L. angustifolius 11 17 5 23 13 5 0 18
L. cosentinii 0 0 0 0 0 0 0 0
L. hispanicus 6 14 14 6 13 3 2 18
L. luteus 2 19 4 15 11 4 2 17
L. micranthus 11 0 11 0 4 1 0 5
a Mid priority spatial gapsb High priority spatial gapsc Gaps in low frequency class, including non-represented ecogeographical categoriesd Gaps in mid-low frequency class
Biodivers Conserv (2012) 21:79–96 85
123
greatest increase corresponded to L. hispanicus (36 of the 80 new accessions belonged
to this species) (Table 2). Comparing the potential and real increase of the CRF-INIA
collection size per species, L. luteus was the species with the lowest collection rate after
populations were detected (*50%).
The species with the greatest number of populations detected and more accessions collected
was L. hispanicus. Lupinus angustifolius generally had slightly lower values than L. hispa-nicus, except in the number of populations detected in priority sites where values were slightly
higher. Populations detected outside priority sites for the target species ranged between 55 and
70% of the total detections or collections and was 100% for L. cosentinii.EPD values for selected priority sites were substantially higher than the corresponding
EPD values for outside priority sites (Table 2). However, the EPD and accession collecting
Fig. 2 Detection of spatial and ecogeographical gaps in the final selection of priority sites forL. angustifolius. a Point map showing CRF-INIA and external presence data, b spatial gaps detected,c ELC map and its 27 categories, d ecogeographical gaps detected, e high potential species richness areasmap and f gaps in areas with 0, 1 and 2 predicted species
86 Biodivers Conserv (2012) 21:79–96
123
Ta
ble
2Q
uan
tita
tive
eval
uat
ion
of
ger
mpla
smex
plo
rati
ons
and
coll
ecti
ng
inse
lect
edpri
ori
tysi
tes
Sp
ecie
sC
RF
-IN
IAco
llec
tio
na
Eff
ecti
ve
size
incr
easi
ng
(%)b
Po
ten
tial
size
incr
easi
ng
(%)c
Nu
mb
ero
fp
rio
rity
site
sd
Nu
mb
ero
fo
ther
site
sv
isit
ede
Po
pu
lati
on
sin
fP
op
ula
tio
ns
ou
tgA
cces
sions
inh
Acc
essi
on
so
uti
EP
Di
(%)j
EP
Do
(%)k
EA
C(%
)l
AB
CD
EF
G
L.
alb
us
24
80
00
10
20
00
0N
/A0
N/A
L. a
ng
ust
ifol
ius
48
96
.18
.81
88
41
72
61
31
79
4.4
31
69
.7
L.
cose
nti
nii
54
04
00
10
20
20
2N
/A2
.0N
/A
L.
his
pa
nic
us
17
52
0.6
26
.31
88
41
53
11
42
28
3.3
36
.97
8.3
L.
lute
us
10
41
1.5
22
.11
78
57
16
57
41
.21
8.8
52
.5
L.
mic
ranth
us
50
05
97
00
00
00
N/A
To
tal
39
75
32
48
aN
um
ber
of
conse
rved
geo
refe
rence
dac
cess
ions
atC
RF
-IN
IAco
llec
tion
bef
ore
2006
bP
erce
nta
ge
of
incr
ease
of
CR
F-I
NIA
coll
ecti
on
size
bas
edon
the
num
ber
of
acce
ssio
ns
coll
ecte
d.
Eff
ecti
ve
size
incr
ease
(%)
=([
F?
G]
91
00
)/A
cP
erce
nta
ge
of
incr
ease
of
CR
F-I
NIA
coll
ecti
on
size
bas
edo
nth
en
um
ber
of
po
pula
tio
ns
det
ecte
d.
Po
ten
tial
size
incr
ease
(%)
=([
D?
E]
91
00
)/A
dN
um
ber
of
pri
ori
tysi
tes
(see
Tab
le1)
eN
um
ber
of
area
so
uts
ide
pri
ori
tysi
tes
defi
ned
for
this
spec
ies
fN
um
ber
of
po
pu
lati
on
sd
etec
ted
insi
de
pri
ori
tysi
tes
defi
ned
for
this
spec
ies
gN
um
ber
of
popula
tions
det
ecte
douts
ide
pri
ori
tysi
tes
defi
ned
for
this
spec
ies
hN
um
ber
of
acce
ssio
ns
(see
d)
coll
ecte
din
side
pri
ori
tysi
tes
defi
ned
for
this
spec
ies
iN
um
ber
of
acce
ssio
ns
(see
d)
coll
ecte
douts
ide
pri
ori
tysi
tes
defi
ned
for
this
spec
ies
jE
PD
bas
edo
nd
etec
tio
ns
insi
de
pri
ori
tysi
tes.
EP
Di
=(D
/B)
91
00
kE
PD
bas
edo
nd
etec
tio
ns
ou
tsid
ep
rio
rity
site
s.E
PD
o=
(E/C
)9
10
0l
EA
C(s
eed
)in
rela
tio
nto
po
pula
tio
ns
fou
nd
insi
de
pri
ori
tysi
tes.
EA
C=
([F
?G
]/[D
?E
])9
10
0
Biodivers Conserv (2012) 21:79–96 87
123
varied among species. For both EPD and EAC, the species with the lowest efficiency
values was L. luteus. In our collecting activities GCI was 0.78.
Qualitative evaluation
Results about qualitative improvement of the ER are shown in the Table 3 and Fig. 4. The
analysis only considers the three widely collected species (L. angustifolius, L. hispanicus
Fig. 3 Sites selected forprospection and results ofcollecting expeditions.a Municipalities where selectedpriority sites were located,b Lupinus populations detected inthe collecting expeditions andc successful seed collecting indetected populations
88 Biodivers Conserv (2012) 21:79–96
123
Table 3 Qualitative evaluation results by species
Species Non-representedcategoriesb
Frequency classesc
Low Mid-low Mid-high High
L. angustifolius
Number of accessions in CRF-INIA collectiona 0 0 56 114 319
Effective size increase (%)a N/A N/A 17.9 6.1 3.1
Potential size increase (%)a N/A N/A 23.2 12.3 3.8
Number of priority sitesa 5 5 13 0 0
Number of other sites visiteda 10 10 19 19 36
Populations ina 4 4 13 0 0
Populations outa 0 0 0 14 12
EPDi (%)a 80 80 100 N/A N/A
EPDo (%)a 0 0 0 73.7 33.3
Accessions ina 3 3 10 0 0
Accessions outa 0 0 0 7 10
EAC (%)a 75 75 76.9 50 83.3
L. hispanicus
CRF-INIA collectiona 0 2 24 30 119
Effective size increase (%)a N/A 150 45.8 13.3 15.1
Potential size increase (%)a N/A 200 45.8 26.7 19.3
Number of priority sitesa 4 4 14 0 0
Number of other sites visiteda 10 12 11 19 42
Populations ina 4 4 11 0 0
Populations outa 0 0 0 8 23
EPDi (%)a 100 100 78.6 N/A N/A
EPDo (%)a 0 0 0 42.1 54.8
Accessions ina 3 3 11 0 0
Accessions outa 0 0 0 4 18
EAC (%)a 75 75 100 50 78.3
L. luteus
CRF-INIA collectiona 0 0 3 20 81
Effective size increase (%)a N/A N/A 66.7 15 4.9
Potential size increase (%)a N/A N/A 100 20 12.3
Number of priority sitesa 4 4 13 0 0
Number of other sites visiteda 14 14 12 36 23
Populations ina 4 4 3 0 0
Populations outa 1 2 0 4 10
EPDi (%)a 100 100 23.1 N/A N/A
EPDo (%)a 7.1 14.3 0 11.1 43.5
Accessions ina 3 3 2 0 0
Accessions outa 0 0 0 3 4
Biodivers Conserv (2012) 21:79–96 89
123
and L. luteus). The case of L. cosentinii is relatively simple, because the two populations
detected and collected had environments not represented in the collection, both out of
priority sites. In fact, for this species there was not any priority site predefined.
For the three species considered, a relevant qualitative improvement was achieved.
Forty-one percent of the new accessions were collected in low and mid-low frequency or
not represented categories in the CRF-INIA collection.
All populations detected and collected in selected priority sites were classified in non-
represented, low and mid-low frequency classes, while those detected and collected out of
priority sites were classified mostly in mid-high and high frequency classes.
The higher increases in the collection size took place in the non-represented, low and
mid-low frequency classes. The highest differences between potential and effective size
increase were observed in L. luteus.
The efficiency in seed collecting once the population had been detected varied
according to the species and the frequency class, but no pattern can be distinguished. The
overall ratio of populations detected summarizing both priority sites and surroundings for
all species showed higher values in the high frequency classes (39.5%), followed by mid-
low (23.7%), mid-high (22.8%) and low (14%) frequency classes. This efficiency in non-
represented categories reached 13.2%. The ratio of accessions collected in each frequency
class was similar to the ratio of populations detected. In this case the highest value was also
for the high frequency class (40%), followed by mid-low (28.8%), mid-high (17.5%) and
low (13.8%) frequency classes. The proportion of accessions collected for non-represented
categories was identical to that of the low frequency class.
For the three species considered, low and mid-low frequency classes in CRF-INIA
collection (before 2006) represented only 11.1% of the accessions conserved. In contrast,
the ratios of populations detected and accessions collected that belonged to non-repre-
sented, low and mid-low frequency classes were 37.8 and 42.5%, showing a clear effort in
collecting in under-represented or non-represented ecogeographical categories.
Finally, all (three) ecogeographical categories from ELC map non-represented in L.angustifolius CRF-INIA collection until 2006 were represented with at least one accession
thanks to the collection missions. Similarly, three of the four and two of the seven non-
represented categories for L. hispanicus and L. luteus, respectively, are now represented
with the new accessions obtained in the collection missions.
Table 3 continued
Species Non-representedcategoriesb
Frequency classesc
Low Mid-low Mid-high High
EAC (%)a 60 50 66.7 75 40
Qualitative results are shown in categories classified by their frequencies. Only species with defined prioritysites and detected populations or sampled accessions are includeda See Table 2b Collecting results for ecogeographical categories from the ELC map not represented in the CRF-INIAcollection until 2006c Classification of ecogeographical categories according to accessions frequency in the CRF-INIA collec-tion until 2006. The low frequency class does not include non-represented ecogeographical categories
90 Biodivers Conserv (2012) 21:79–96
123
Discussion
The use of species distribution models and ecogeographical maps
Species distribution models have previously been used for collecting plant genetic
resources (Segura et al. 2003; Jarvis et al. 2005). Positive results have been reported in
terms of number of populations detected, since the models predict potential collecting sites
based on the most favorable environments for the target species. However, in terms of ER,
the exclusive use of species distribution models for germplasm collecting may lead to
redundancy. Because the aim of species distribution models is to predict a species’
Fig. 4 Improvement in ecogeographical representativeness by category in three of six Lupinus species.Grey bars represent the former CRF-INIA germplasm distribution and black bars represent the newcollected accessions distribution. Ecogeographical categories are labeled according to frequency classesas follows: L low, ML mid-low, MH mid-high and H high frequency. a L. angustifolius, b L. hispanicusand c L. luteus
Biodivers Conserv (2012) 21:79–96 91
123
occurrence based on a ‘‘minimum habitat quality threshold below which it is presumed the
species will not occur’’ (Larson et al. 2008), they tend to guide collectors to the most
preferred habitats for the species. In our case, this would mean collecting germplasm in
populations in the mid-high and high frequency classes, that is, the most frequent envi-
ronments. Thus, if collecting expeditions had only been guided by the species distribution
models, the more frequent environments where the species occurs would have been over-
represented in the CRF-INIA Lupinus collection. In contrast, filtering potential collecting
sites with the use of ELC maps avoided over-representation.
The inclusion in the CRF-INIA collection of accessions from low and non-represented
ecogeographical categories, equivalent to marginal environments of the species’ range,
may provide a very important source of traits (genes) related to abiotic tolerance in plant
breeding (Parra-Quijano et al. 2008). Nevertheless, collecting in rare or marginal envi-
ronments has some disadvantages, such as reducing the efficiency of accession collecting
due to the scarce number of individuals per population and/or low production of seeds.
Loss of fitness of plant individuals in marginal habitats is evident when we compared
populations belonging to the high and low frequency classes (Fig. 5).
Quantitative and qualitative evaluation of representativeness
Since the aim of collecting explorations was to improve the ER of an existing germplasm
collection, better results were expected in the qualitative than in the quantitative evalua-
tion. Several populations from external data were not pursued to avoid ecogeographical
redundancy. These restrictions reduced quantitative efficiency. In spite of these limitations,
if all detected populations had been collected, the size of the CRF-INIA georeferenced
Lupinus collection would have increased over 10%. Such an increase can be considered
acceptable for existing collections if we compare these results to those of ‘‘standard’’
Fig. 5 Phenotypes of typical L. angustifolius plants from two populations found in contrastingecogeographical categories. a High frequency class (Carrascal de Barregas, Salamanca, 40�5805900 N,5�5005800W) and b non-represented category (Zarzalejo, Madrid, 40�3301500N, 4�110200W)
92 Biodivers Conserv (2012) 21:79–96
123
collecting activities reported in other gene banks. Thus, Lane et al. (2000) reported a 7%
increase after collecting expeditions of white clover (Trifolium repens).
Another comparable parameter among collecting projects is the GCI. In a multi-forage
species exploration where Lupinus was one of several target species, Lazaro et al. (2002),
visited 65 sites in Spain and collected 12 Lupinus accessions obtaining a GCI of 0.18. The
differences between this value and our GCI (0.78) can be attributed both to the effec-
tiveness of the spatial gap analysis and to the fact that our collecting expeditions only
focused on Lupinus species.
The qualitative evaluation showed ‘‘balanced’’ collecting results between low and high
represented environments in the CRF-INIA collection. Obviously, collecting accessions
outside of the selected priority sites favored the highly represented environments.
Similarly, high EPD values were expected in the low and mid-low frequency classes,
since efforts focused on priority sites which corresponded to low represented environ-
ments. Lower efficiency values (both EPD and EAC) in L. luteus could be related to the
issues discussed in the next section of this discussion. Although we have no data on genetic
erosion for cultivated populations of L. luteus, there is enough evidence of risk in wild
populations (Lazaro et al. 2002). Further studies and initiatives in in situ conservation
might be necessaries if the threats to this species in natural populations are confirmed.
The methodology in practice
Detecting spatial or ecogeographical gaps and using species distribution models to improve
the representativeness of ex situ gene banks are new methodological approaches that can
provide reliable results (Parra-Quijano et al. 2008; Ramirez-Villegas et al. 2010). However,
the expected results may be overly optimistic, since biological and field expedition con-
ditions can significantly reduce initial expectations, in terms of detecting populations and
collecting accessions. Thus, in order to obtain a more realistic assessment of any collecting
method based on gap analysis, it is necessary to implement the method in real field
collections and to carry out evaluations and validations from its results.
The success of collecting activities is subject to many factors, some of which can be
difficult to control. In our case, climatic conditions affect flowering time and seed set.
These traits are important in detecting and collecting Lupinus populations. Lupinus plants
are difficult to locate without flowers. Thus, although the date of the first expeditions was
adjusted to climatic conditions, it was not possible to determine a single date to detect
populations of all species at the same time.
Human activities also have direct effect on seed collecting once the population has been
detected. For instance, entire populations have been eliminated after their detection due to
road edge cleaning activities. In addition, some of these wild Lupinus species are con-
sidered to be weeds and farmers tend to eliminate them from their fields. Policies and
legislation to protect this type of germplasm should be developed to national and global
scales to avoid these issues (Iriondo et al. 2008).
The difference between the number of populations detected and collected can mainly be
explained by the natural and human factors mentioned above. These factors had a greater
effect on L. luteus than the rest of species, probably because this species is more sensitive
to changes in environmental conditions or habitat alteration.
As expected, the simpatric distribution of Lupinus species across Iberian Peninsula
(Cowling et al. 1998) allowed multiple species to be collected per site. In this sense, the
map of potential Lupinus species richness used in the proposed methodology took
advantage of this feature.
Biodivers Conserv (2012) 21:79–96 93
123
Different reasons may explain why no seeds were collected for L. albus and L. mi-cranthus. Lupinus albus is found in Iberian Peninsula exclusively as a crop species
(Castroviejo and Pascual 1999). Since crop species have a low representation in herbaria
data, no priority sites were obtained for L. albus. Additionally, the sowing of crop species
depends on several agricultural and anthropic factors. In some sites local farmers con-
firmed L. albus occurrence, although no plants were detected in the field and seeds were
not available. Furthermore, the cultivation of this species has been decreasing in Spain in
the past decades (Simpson and McGibbon 1982).
In the case of L. micranthus, no populations were found even though five priority sites
were defined for this species. In contrast, no priority sites were defined for L. cosentinii, but
two populations of this species were detected in the priority sites defined for L. micranthus.These contradictory results could be due to incorrect taxonomic identification of the
external source. Both species have similar flowers and overlapping distribution areas in
the Iberian Peninsula (Pascual 1986; Castroviejo and Pascual 1999). Therefore, the two
priority sites defined for L. micranthus may in fact correspond to L. cosentinii. In any case,
the effectiveness of the collecting methodology seems to be lower in rare species, at least
in quantitative terms.
For prospective users of this methodology, it is important to point out its flexibility
when applied to different species or territories. For instance, the set of ecogeographical
variables can be adjusted according to their availability or depending on the adaptative
features of each species. In this sense, for less proficient GIS users, creating customized
ELC maps may be an obstacle, but other already available ecosystem or ecological maps
of the territory of interest may also be used. The other techniques and tools used can
be applied with available user-friendly software, such as DIVA-GIS for gap analysis
(http://www.diva-gis.org), or MAXENT (Phillips et al. 2006) for species distribution
modeling (http://www.cs.princeton.edu/*schapire/maxent/).
Cost-benefit balance of the collecting methodology
Filling gaps in germplasm collections is probably more expensive than generalized
collecting projects in terms of cost per collected accession. However, the joint use of GIS
tools, gap analysis, species distribution models, and ELC maps can help to reduce their
cost, as it allows to increase the EPD. Another point to bear in mind is that the proposed
methodology is based on studies of ER. Thus, it does not require expensive infrastructure
or highly-qualified personal (in contrast to the genotypic or phenotypic approaches), which
reduces the costs of its implementation. Finally, it is important to note that the ultimate
goal of ex situ collections is to represent a species’ natural variation and when this is not
met, the purpose and use of the collections can be significantly curtailed. In this sense, the
use of ELC maps and gap analysis contribute to increase the value of a collection, as new
accessions from marginal environments (probably with important traits related to abiotic
tolerance) could be added.
Conclusions
The methodology proposed here offers highly efficient collecting values combined with
a successful improvement in representativeness. The achievement of a complete ER of the
L. angustifolius collection in the CRF-INIA gene bank and the substantial improvement
of ER in the case of L. hispanicus and L. luteus are evidence of the feasibility of this type
94 Biodivers Conserv (2012) 21:79–96
123
of collecting in existing germplasm collections. Forthcoming developments and
improvements in GIS software, environmental information, accession and population
presence data and modeling techniques provide additional expectations on further
increasing the efficiency of collecting expeditions.
Acknowledgments This work was funded by INIA (Ministry of Science and Innovation) project RF2004-00016-00-00. The authors thank all CRF-INIA personnel and UPM Plant Biology Department who sup-ported collecting activities. We also thank Lori de Hond for her linguistic assistance.
References
Castroviejo S, Pascual H (1999) Lupinus. In: Talavera S, Aedo C, Castroviejo S, Romero Zarco C, Saez L,Salgueiro FJ, Velayos M (eds) Flora iberica: plantas vasculares de la Penınsula Iberica e Islas Baleares,vol 7. Consejo Superior de Investigaciones Cientıficas, Madrid, pp 251–260
Cowling WA, Buirchell BJ, Tapia ME (1998) Lupin, Lupinus L. Institute of Plant Genetics and Crop PlantResearch, Gatersleben; International Plant Genetic Resources Institute, Rome
FAO (2010) The second report on the state of the world‘s plant genetic resources for food and agriculture.FAO, Rome
Fielding AH, Bell JF (1997) A review of methods for the assessment of prediction errors in conservationpresence/absence models. Environ Conserv 24:38–49
French RJ, Sweetingham MW, Shea GG (2001) A comparison of the adaptation of yellow lupin (Lupinusluteus L.) and narrow-leafed lupin (L. angustifolius L.) to acid sandplain soils in low rainfall agri-cultural areas of Western Australia. Aust J Agric Res 52:945–954
Gladstones JS (1980) Recent developments in the understanding, improvement, and use of Lupinus. In:Summerfield RJ, Bunting AH (eds) Advances in legume science. Royal Botanical Gardens, Kew,pp 603–611
Greene SL, Hart TC (1999) Implementing a geographic analysis in germplasm conservation. In: Greene SL,Guarino L (eds) Linking genetic resources and geography: emerging strategies for conserving andusing crop biodiversity. American Society of Agronomy; Crop Science Society of America, Madison,pp 25–38
Guarino L, Jarvis A, Hijmans R, Maxted N (2002) Geographic information systems (GIS) and the con-servation and use of plant genetic resources. In: Engels JMM, Ramanatha Rao V, Brown AHD,Jackson MT (eds) Managing plant genetic diversity. CAB International, Wallingford, pp 387–404
Hamblin J, Delane R, Bishop A, Gladstones J (1986) Yield potential of reduced-branching lupins (Lupinusangustifolius) on sandy soils in a short season environment. Aust J Agric Res 37:611–620
Hijmans RJ, Schreuder M, De la Cruz J, Guarino L (1999) Using GIS to check co-ordinates of genebankaccessions. Genet Resour Crop Evol 46:291–296
Hijmans RJ, Garrett KA, Huaman Z, Zhang DP, Schreuder M, Bonierbale M (2000) Assessing the geo-graphic representativeness of genebank collections: the case of Bolivian wild potatoes. Conserv Biol14:1755–1765
Hijmans RJ, Guarino L, Cruz M, Rojas E (2001) Computer tools for spatial analysis of plant geneticresources data: 1. DIVA-GIS. Plant Genet Resour Newsl 127:15–19
Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A (2005) Very high resolution interpolated climatesurfaces for global land areas. Int J Climatol 25:1965–1978
Huyghe C (1997) White lupin (Lupinus albus L.). Field Crops Res 53:147–160Instituto Geografico Nacional (1992) Atlas nacional de Espana, seccion II, grupo 7. Edafologıa Instituto
Geografico Nacional, MadridIriondo JM, Dulloo ME, Maxted N, Laguna E, Engels JMM, Maggioni L (2008) Final considerations for the
in situ conservation of plant genetic diversity. In: Iriondo JM, Maxted N, Dulloo ME (eds) Conservingplant genetic diversity in protected areas. CAB International, Wallingford, pp 182–202
Jarvis A, Williams K, Williams D, Guarino L, Caballero PJ, Mottram G (2005) Use of GIS for optimizing acollecting mission for a rare wild pepper (Capsicum flexuosum Sendtn.) in Paraguay. Genet ResourCrop Evol 52:671–682
Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics33:159–174
Lane LA, Ayres JF, Lovett JV (2000) A white clover (Trifolium repens L.) ecotype collection from northernNew South Wales, Australia. Plant Genet Resour Newsl 121:10–14
Biodivers Conserv (2012) 21:79–96 95
123
Larson MA, Millspaugh JJ, Thompson FR (2008) A review of methods for quantifying wildlife habitat inlarge landscapes. In: Millspaugh JJ, Thompson FR (eds) Models for planning wildlife conservation inlarge landscapes. Academic Press, Burlington, pp 225–250
Lazaro A, Campbell M, Francis C, Robertson L, de la Cuadra C (2002) Collecting biodiversity of importantforage legumes in Spain. Plant Genet Resour Newsl 129:11–16
Maxted N, Dulloo E, Ford-Lloyd BV, Iriondo JM, Jarvis A (2008) Gap analysis: a tool for complementarygenetic conservation assessment. Divers Distrib 14:1018–1030
Parra-Quijano M, Draper D, Iriondo JM (2007) GIS-based evaluation of the in situ conservation of a cropwild relative: the case of Spanish lupins. Bocconea 21:105–116
Parra-Quijano M, Draper D, Torres E, Iriondo JM (2008) Ecogeographical representativeness in crop wildrelative ex situ collections. In: Maxted N, Ford-Lloyd BV, Kell SP, Iriondo JM, Dulloo ME, Turok J(eds) Crop wild relative conservation and use. CAB International, Wallingford, pp 249–273
Parra-Quijano M, Iriondo JM, Torres E (2011) Ecogeographical land characterization maps as a tool forassessing plant adaptation and their implications in agrobiodiversity studies. Genet Resour Crop Evol.doi:10.1007/s10722-011-9676-7
Pascual H (1986) Altramuces de la Penınsula Iberica e Islas Baleares. Taxonomıa, area, autoecologıa,aprovechamiento y aspectos agronomicos. Instituto Nacional de Investigaciones Agrarias, Madrid
Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species geographic distri-butions. Ecol Model 190:231–259
Rabus B, Eineder M, Roth A, Bamler R (2003) The shuttle radar topography mission–a new class of digitalelevation models acquired by spaceborne radar. ISPRS-J Photogramm Remote Sens 57:241–262
Rahman MS, Gladstones JS (1974) Effects of temperature and photoperiod on flowering and yield com-ponents of lupin genotypes in the field. Aust J Exp Agric Anim Husb 14:205–213
Ramirez-Villegas J, Khoury C, Jarvis A, Debouck D, Guarino L (2010) A gap analysis methodologyfor collecting crop genepools: a case study with Phaseolus beans. PLoS ONE 5:e13497. doi:10.1371/journal.pone.0013497
Segura S, Coppens d’Eeckenbrugge G, Lopez L, Grum M, Guarino L (2003) Mapping the potential dis-tribution of five species of Passiflora in Andean countries. Genet Resour Crop Evol 50:555–566
Simpson MJA, McGibbon R (1982) White lupin in cultivation in Iberia. Econ Bot 36:442–446SPSS (2003) SPSS base 12.0 user’s guide. SPSS Inc, ChicagoSujak A, Kotlarz A, Strobel W (2006) Compositional and nutritional evaluation of several lupin seeds.
Food Chem 98:711–719
96 Biodivers Conserv (2012) 21:79–96
123